modules/private/monitoring/objects_common.nix

   1 { hostFQDN
   2 , hostName
   3 , processWarn ? "250"
   4 , processAlert ? "400"
   5 , loadWarn ? "8.0"
   6 , loadAlert ? "10.0"
   7 , mdadm
   8 , sudo
   9 , master
  10 , lib
  11 , ...
  12 }:
  13 let
  14   defaultPassiveInfo = {
  15     filter = lib.attrsets.filterAttrs
  16       (k: v: builtins.elem k ["service_description"] || builtins.substring 0 1 k == "_");
  17     use = "external-passive-service";
  18     freshness_threshold = "450";
  19     retry_interval = "1";
  20     servicegroups = "webstatus-resources";
  21     host_name = hostFQDN;
  22   };
  23 in
  24 {
  25   host = {
  26     "${hostFQDN}" = {
  27       alias = hostFQDN;
  28       address = hostFQDN;
  29       use = "linux-server";
  30       hostgroups = "webstatus-hosts";
  31       _webstatus_name = hostName;
  32     };
  33   };
  34   service = [
  35     {
  36       passiveInfo = defaultPassiveInfo;
  37       service_description = "Size on root partition";
  38       use = "local-service";
  39       check_command = ["check_local_disk" "20%" "10%" "/"];
  40     }
  41     {
  42       passiveInfo = defaultPassiveInfo;
  43       service_description = "Total number of process";
  44       use = "local-service";
  45       check_command = [
  46         "check_local_procs"
  47         processWarn
  48         processAlert
  49         "RSZDT"
  50       ];
  51     }
  52     {
  53       passiveInfo = defaultPassiveInfo;
  54       service_description = "Average load";
  55       use = "local-service";
  56       check_command = [
  57         "check_local_load"
  58         "${loadWarn},${loadWarn},${loadWarn}"
  59         "${loadAlert},${loadAlert},${loadAlert}"
  60       ];
  61     }
  62     {
  63       passiveInfo = defaultPassiveInfo;
  64       service_description = "Swap usage";
  65       use = "local-service";
  66       check_command = ["check_local_swap" "20" "10"];
  67     }
  68     {
  69       passiveInfo = defaultPassiveInfo;
  70       service_description = "Memory usage";
  71       use = "local-service";
  72       check_command = ["check_memory" "80" "90"];
  73     }
  74     {
  75       passiveInfo = defaultPassiveInfo;
  76       service_description = "NTP is activated and working";
  77       use = "local-service";
  78       check_command = ["check_ntp"];
  79     }
  80     {
  81       passiveInfo = defaultPassiveInfo;
  82       service_description = "No mdadm array is degraded";
  83       use = "local-service";
  84       check_command = [
  85         "check_command_output"
  86         "${mdadm}/bin/mdadm --monitor --scan -1"
  87         "^$"
  88         "-s 0 -r root"
  89       ];
  90     }
  91   ];
  92   command = {
  93     check_dns = "$USER1$/check_dns -H $ARG1$ -s $HOSTADDRESS$ $ARG2$";
  94     check_emails = "$USER2$/check_emails -H $HOSTADDRESS$ -i $USER203$ -l $ARG1$ -p $ARG2$ -s $ARG3$ -f $ARG4$";
  95     check_emails_local = "$USER2$/check_emails -H $HOSTADDRESS$ -n $ARG1$ -r $ADMINEMAIL$ -s $ARG2$ -f $ARG3$";
  96     check_eriomem = "$USER2$/check_eriomem $USER208$";
  97     check_eriomem_age = "$USER2$/check_eriomem_age $ARG1$";
  98     check_external_dns = "$USER1$/check_dns -H $ARG2$ -s $ARG1$ $ARG3$";
  99     check_ftp_database = "$USER2$/check_ftp_database";
 100     check_git = "$USER2$/check_git $USER203$";
 101     check_http = "$USER1$/check_http --sni -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 102     check_https = "$USER1$/check_http --sni --ssl -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 103     check_https_auth = "$USER1$/check_http --sni --ssl -a \"$USER202$\" -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 104     check_https_certificate = "$USER1$/check_http --sni --ssl -H \"$ARG1$\" -C 21,15";
 105     check_https_code = "$USER1$/check_http --sni --ssl -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -e \"$ARG3$\" -r \"$ARG4$\"";
 106     check_imap_connection = "$USER2$/check_imap_connection -u \"$USER204$\" -p \"$USER205$\" -H \"imap.immae.eu:143\"";
 107     check_local_disk = "$USER1$/check_disk -w $ARG1$ -c $ARG2$ -p $ARG3$";
 108     check_local_procs = "$USER1$/check_procs -w $ARG1$ -c $ARG2$ -s $ARG3$";
 109     check_local_load = "$USER1$/check_load -w $ARG1$ -c $ARG2$";
 110     check_local_swap = "$USER1$/check_swap -n ok -w $ARG1$ -c $ARG2$";
 111     check_memory = "$USER2$/check_mem.sh -w $ARG1$ -c $ARG2$";
 112     check_command_match = "$USER2$/check_command -c \"$ARG1$\" -C \"$ARG2$\" $ARG3$";
 113     check_command_output = "$USER2$/check_command -c \"$ARG1$\" -s 0 -o \"$ARG2$\" $ARG3$";
 114     check_command_status = "$USER2$/check_command -c \"$ARG1$\" -s \"$ARG2$\" $ARG3$";
 115     check_ntp = "$USER1$/check_ntp_time -t 30 -q -H 0.arch.pool.ntp.org";
 116     check_mailq = "$USER1$/check_mailq -s -w 1 -c 2";
 117     check_mysql_replication = "${sudo} -u mysql $USER2$/check_mysql_replication \"$ARG1$\" \"$ARG2$\"";
 118     check_postgresql_replication = "${sudo} -u postgres $USER2$/check_postgres_replication \"$ARG1$\" \"$ARG2$\" \"$ARG3$\"";
 119     check_openldap_replication = "${sudo} -u openldap $USER2$/check_openldap_replication \"$ARG1$\" \"$ARG2$\" \"$ARG3$\" \"$ARG4$\" \"$ARG5$\"";
 120     check_ovh_sms = "$USER2$/check_ovh_sms \"$USER209$\"";
 121     check_redis_replication = "${sudo} -u redis $USER2$/check_redis_replication \"$ARG1$\"";
 122     check_smtp = "$USER1$/check_smtp -H $HOSTADDRESS$ -p 25 -S -D 21,15";
 123     check_tcp = "$USER1$/check_tcp -H $HOSTADDRESS$ -p $ARG1$ -e \"$ARG2$\" -Mcrit";
 124     check_tcp_ssl = "$USER1$/check_tcp -H $HOSTADDRESS$ -p $ARG1$ -S -D 21,15";
 125
 126     check_host_alive = "$USER1$/check_ping -H $HOSTADDRESS$ -w 3000.0,80% -c 5000.0,100% -p 5";
 127     check_last_file_date = "${sudo} -u \"$ARG3$\" $USER2$/check_last_file_date \"$ARG1$\" \"$ARG2$\"";
 128     check_ok = "$USER1$/check_dummy 0 \"Dummy OK\"";
 129     check_critical = "$USER1$/check_dummy 2 \"Dummy CRITICAL\"";
 130
 131     # $OVE is to force naemon to run via shell instead of execve which fails here
 132     notify-host-by-email = "ADMINEMAIL=\"$ADMINEMAIL$\" SERVICENOTIFICATIONID=\"$SERVICENOTIFICATIONID$\" HOSTSTATE=\"$HOSTSTATE$\" HOSTOUTPUT=\"$HOSTOUTPUT$\" $USER2$/notify_by_email host \"$NOTIFICATIONTYPE$\" \"$HOSTALIAS$\" \"$LONGDATETIME$\" \"$CONTACTEMAIL$\" $OVE";
 133     # $OVE is to force naemon to run via shell instead of execve which fails here
 134     notify-service-by-email = "ADMINEMAIL=\"$ADMINEMAIL$\" SERVICENOTIFICATIONID=\"$SERVICENOTIFICATIONID$\" SERVICEDESC=\"$SERVICEDESC$\" SERVICESTATE=\"$SERVICESTATE$\" SERVICEOUTPUT=\"$SERVICEOUTPUT$\" $USER2$/notify_by_email service \"$NOTIFICATIONTYPE$\" \"$HOSTALIAS$\" \"$LONGDATETIME$\" \"$CONTACTEMAIL$\" $OVE";
 135     notify-by-slack = "HOST=\"$HOSTALIAS$\" SERVICESTATE=\"$SERVICESTATE$\" SERVICEDESC=\"$SERVICEDESC$\" SERVICEOUTPUT=\"$SERVICEOUTPUT$\" $USER2$/notify_by_slack \"$ARG1$\" \"$ARG2$\"";
 136
 137     notify-master = "$USER2$/send_nrdp.sh -u \"$USER200$\" -t \"$USER201$\" -H \"$HOSTADDRESS$\" -s \"$SERVICEDESC$\" -S \"$SERVICESTATEID$\" -o \"$SERVICEOUTPUT$ | $SERVICEPERFDATA$\"";
 138   };
 139   timeperiod = {
 140     "24x7" = {
 141       alias = "24 Hours A Day, 7 Days A Week";
 142       monday = "00:00-24:00";
 143       tuesday = "00:00-24:00";
 144       wednesday = "00:00-24:00";
 145       thursday = "00:00-24:00";
 146       friday = "00:00-24:00";
 147       saturday = "00:00-24:00";
 148       sunday = "00:00-24:00";
 149     };
 150   };
 151   servicegroup = {
 152     webstatus-webapps = { alias = "Web applications"; };
 153     webstatus-websites = { alias = "Personal websites"; };
 154     webstatus-ssl = { alias = "SSL certificates"; };
 155     webstatus-dns = { alias = "DNS resolution"; };
 156     webstatus-remote-services = { alias = "Other remote services"; };
 157     webstatus-local-services = { alias = "Other local services"; };
 158     webstatus-email = { alias = "E-mail services"; };
 159     webstatus-resources = { alias = "Local resources"; };
 160     webstatus-databases = { alias = "Databases resources"; };
 161     webstatus-backup = { alias = "Backup resources"; };
 162   };
 163   hostgroup = {
 164     webstatus-hosts = { alias = "Hosts"; };
 165   };
 166   contactgroup = {
 167     admins = { alias = "Naemon Administrators"; };
 168   };
 169   templates = {
 170     service = {
 171       generic-service = {
 172         active_checks_enabled = "1";
 173         check_freshness = "0";
 174         check_interval = "10";
 175         check_period = "24x7";
 176         contact_groups = "admins";
 177         event_handler_enabled = "1";
 178         flap_detection_enabled = "1";
 179         is_volatile = "0";
 180         max_check_attempts = "3";
 181         notification_interval = "60";
 182         notification_options = "w,u,c,r,f,s";
 183         notification_period = "24x7";
 184         notifications_enabled = if master then "1" else "0";
 185         obsess_over_service = "1";
 186         passive_checks_enabled = "1";
 187         process_perf_data = "1";
 188         retain_nonstatus_information = "1";
 189         retain_status_information = "1";
 190         retry_interval = "2";
 191       };
 192       local-service = {
 193         use = "generic-service";
 194         host_name = hostFQDN;
 195         check_interval = "5";
 196         max_check_attempts = "4";
 197         retry_interval = "1";
 198         servicegroups = "webstatus-resources";
 199       };
 200       external-service = {
 201         use = "generic-service";
 202         check_interval = "5";
 203         max_check_attempts = "4";
 204         retry_interval = "1";
 205       };
 206       web-service = {
 207         use = "generic-service";
 208         check_interval = "20";
 209         max_check_attempts = "2";
 210         retry_interval = "1";
 211       };
 212       external-web-service = {
 213         use = "generic-service";
 214         check_interval = "20";
 215         max_check_attempts = "2";
 216         retry_interval = "1";
 217       };
 218       mail-service = {
 219         use = "generic-service";
 220         check_interval = "15";
 221         max_check_attempts = "1";
 222         retry_interval = "1";
 223       };
 224       dns-service = {
 225         use = "generic-service";
 226         check_interval = "120";
 227         notification_interval = "120";
 228         max_check_attempts = "5";
 229         retry_interval = "5";
 230       };
 231     };
 232     # No contact, we go through master
 233     contact = {
 234       generic-contact = {
 235         host_notification_commands = "notify-host-by-email";
 236         host_notification_options = "d,u,r,f,s";
 237         host_notification_period = "24x7";
 238         service_notification_commands = "notify-service-by-email";
 239         service_notification_options = "w,u,c,r,f,s";
 240         service_notification_period = "24x7";
 241       };
 242     };
 243     host = {
 244       generic-host = {
 245         event_handler_enabled = "1";
 246         flap_detection_enabled = "1";
 247         notification_period = "24x7";
 248         notifications_enabled = "1";
 249         process_perf_data = "1";
 250         retain_nonstatus_information = "1";
 251         retain_status_information = "1";
 252       };
 253       linux-server = {
 254         check_command = "check_host_alive";
 255         check_interval = "5";
 256         check_period = "24x7";
 257         contact_groups = "admins";
 258         max_check_attempts = "10";
 259         notification_interval = "120";
 260         notification_options = "d,u,r,f";
 261         retry_interval = "1";
 262       };
 263     };
 264   };
 265 }